A method and system for evaluating the rock breaking effect of a rotary cutter

Abstract

The application provides a method and system for evaluating the effect of a rolling cutter breaking rock. The method comprises: selecting a first recommended area from a plurality of candidate working areas according to point cloud information; obtaining a second recommended area from a plurality of candidate working areas according to image information and selected rolling cutter parameters; obtaining a working evaluation prediction value according to the coincidence degree of the first recommended area and the second recommended area, and updating the second recommended area according to new image information and new selected rolling cutter parameters when the working evaluation prediction value is lower than a preset value; taking the working evaluation prediction value not lower than the preset value and the corresponding selected rolling cutter parameters as rock breaking decision information; performing a rock breaking test; obtaining inspection information including image data for indicating the geometric characteristics of rock residue after rock breaking, and obtaining an evaluation result of the rolling cutter breaking rock according to the inspection information and the working evaluation prediction value in the rock breaking decision information. The closed-loop logic of multi-modal data fusion-iterative optimization is used in the process of evaluating the effect of the rolling cutter breaking rock.

Description

Technical Field

[0001] This application belongs to the fields of image processing and engineering automation, and in particular relates to a method and system for evaluating the rock-breaking effect of a roller cutter. Background Technology

[0002] With the continuous development of underground space, the excavation difficulty and mechanization level of tunnel engineering are gradually increasing, and full-face tunneling has been rapidly developed and applied. Compared with the traditional drill and blast method, it has advantages such as fast construction speed, high tunnel quality, and less construction disturbance.

[0003] In its corresponding experiments, taking the dynamic rock-breaking efficiency evaluation method disclosed in application number CN202110663955.5 as an example, the rock-breaking efficiency of the drill bit and rock interaction is evaluated through dynamic rock-breaking experiments and energy calculation methods, providing guidance for optimizing drilling parameters and methods, improving drilling efficiency and reducing costs. However, it focuses more on the later energy calculation and lacks multimodal perception of the geometric-mechanical heterogeneity of the rock mass and optimization of the cutter parameters in the pre-experiment stage of rock-breaking tests. That is, the initial setting of the test area and the working conditions of the cutter still relies on experience, which cannot fundamentally reduce the trial and error costs caused by the heterogeneity of the rock mass.

[0004] Based on this, this application provides a method and system for evaluating the rock-breaking effect of a roller cutter, in order to improve related technologies. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this application is to provide a method and system for evaluating the rock-breaking effect of a roller cutter.

[0006] The first aspect of this application proposes a method for evaluating the rock-breaking effect of a rolling cutter, comprising:

[0007] Acquire point cloud information and image information of multiple candidate working areas including the target rock working surface, wherein the size of the cutter placement area of ​​each candidate working area meets the spatial constraints of the roller cutter rock breaking operation; select a first recommended area from the multiple candidate working areas based on the point cloud information; and obtain a second recommended area from the multiple candidate working areas based on the image information and the selected roller cutter parameters.

[0008] The job evaluation prediction value is obtained based on the overlap between the first recommended region and the second recommended region. When the job evaluation prediction value is lower than the preset value, the image information and the selected cutting tool parameters are re-acquired, and the second recommended region is updated based on the new image information and the new selected cutting tool parameters. The job evaluation prediction value is re-judged using the first recommended region and the updated second recommended region until the job evaluation prediction value is not lower than the preset value.

[0009] The predicted value of the operation assessment when it is not lower than the preset value and the corresponding selected cutter parameters are used as rock breaking decision information; an operation strategy is generated based on the selected cutter parameters in the rock breaking decision information and a rock breaking test is conducted on the target rock; inspection information including image data used to indicate the geometric characteristics of the rock debris after rock breaking is obtained, and the evaluation result of cutter rock breaking is obtained based on the inspection information and the predicted value of the operation assessment in the rock breaking decision information.

[0010] Preferably, the acquisition of point cloud information and image information of multiple candidate working areas including the target rock working surface includes:

[0011] Multiple line laser scanners are rotated and positioned in front of the target rock work area to acquire point cloud data covering each candidate work area;

[0012] Simultaneously use multiple industrial cameras to capture images of each candidate work area from different perspectives, register and correct the distortion of the multi-view images, and generate image information.

[0013] The point cloud data and the image information are synchronized in time and space using timestamps and a spatial calibration board to form a one-to-one point cloud-image data pair, which is then stored in the database.

[0014] Preferably, the step of obtaining a second recommended region from multiple candidate working regions based on the image information and the selected hobbing parameters includes:

[0015] The image information and the selected hobbing cutter parameters are input into the target area recommendation model, and the candidate operation area corresponding to the output area range is used as the second recommended area.

[0016] The target region recommendation model is trained from a deep learning model, and the training process of the target region recommendation model includes the following steps:

[0017] Obtain a training set, which includes multiple sets of training data. Each set of training data includes sample image data, sample roller cutter parameters, and annotation data of the target region corresponding to the sample image data and the sample roller cutter parameters. For each set of training data in the training set, perform the following processing:

[0018] The sample image data and sample roller parameters in the training data are input into a preset deep learning model to obtain the prediction data of the corresponding target area.

[0019] Based on the obtained prediction data and labeled data, the model parameters of the deep learning model are updated; it is checked whether the preset training termination condition is met; if yes, the trained deep learning model is used as the target region recommendation model; if no, the deep learning model is trained again using the next training data.

[0020] Preferably, the selected hob parameters are obtained through the following methods:

[0021] Obtain the mapping relationship between preset image information and preset cutter parameters; based on the mapping relationship, obtain the preset cutter parameters corresponding to the preset image information with the highest correlation to the image information, and use them as the selected cutter parameters;

[0022] When the predicted value of the job evaluation is lower than the preset value, the preset cutter parameter corresponding to the currently selected cutter parameter is masked from the mapping relationship. The preset cutter parameter corresponding to the preset image information with the highest correlation with the image information is obtained according to the mapping relationship and used as the new selected cutter parameter.

[0023] Preferably, the step of selecting a first recommended region from multiple candidate operation regions based on the point cloud information includes:

[0024] For each candidate work area, the normal vector is calculated from the point cloud, and the area with a local curvature change rate greater than a preset threshold is extracted as a candidate area for high stress concentration.

[0025] Based on curvature similarity and spatial continuity, a region growing algorithm is used to aggregate candidate regions of high stress concentration into several continuous regions, which are then used as the first recommended regions, and their boundary coordinates are output.

[0026] Preferably, obtaining the job evaluation prediction value based on the overlap between the first recommended region and the second recommended region includes:

[0027] Project the first recommended region and the second recommended region onto the same reference plane coordinate system, and calculate the ratio of their intersection area to their union area as the degree of overlap;

[0028] If the overlap is not less than its corresponding preset overlap, the operation evaluation prediction value is obtained based on the overlap and the roller rock breaking efficiency coefficient; if the overlap is less than its corresponding preset overlap, the operation evaluation prediction value is obtained based on the overlap, correction coefficient and roller rock breaking efficiency coefficient.

[0029] The efficiency coefficient of the cutter rock breaking is obtained by referring to a table from the cutter head rotation speed, penetration depth and uniaxial compressive strength of the rock in the cutter parameters.

[0030] Preferably, obtaining the evaluation result of the roller cutter rock breaking based on the inspection information and the operation evaluation prediction value in the rock breaking decision information includes:

[0031] Based on the inspection information, evaluation data is obtained through the rock breaking evaluation model; based on the evaluation data and the operation evaluation prediction value, the evaluation result of the roller cutter rock breaking is obtained.

[0032] A second aspect of this application proposes an evaluation system for the rock-breaking effect of a rolling cutter, comprising:

[0033] The information acquisition module is used to acquire point cloud information and image information of multiple candidate working areas including the target rock working face, wherein the size of the cutter placement area of ​​each candidate working area meets the spatial constraints of the roller cutter rock breaking operation; based on the point cloud information, a first recommended area is selected from the multiple candidate working areas; based on the image information and the selected roller cutter parameters, a second recommended area is acquired from the multiple candidate working areas.

[0034] The region assessment module is used to obtain the job assessment prediction value based on the overlap between the first recommended region and the second recommended region. When the job assessment prediction value is lower than the preset value, the image information and the selected hobbing parameters are re-acquired, and the second recommended region is updated based on the new image information and the new selected hobbing parameters. The job assessment prediction value is re-judged using the first recommended region and the updated second recommended region until the job assessment prediction value is not lower than the preset value.

[0035] The result acquisition module is used to take the operation evaluation prediction value when it is not lower than a preset value and its corresponding selected cutter parameters as rock breaking decision information; generate an operation strategy based on the selected cutter parameters in the rock breaking decision information and conduct a rock breaking test on the target rock; acquire inspection information including image data used to indicate the geometric characteristics of rock debris after rock breaking; and obtain the evaluation result of cutter rock breaking based on the inspection information and the operation evaluation prediction value in the rock breaking decision information.

[0036] A third aspect of this application provides an electronic device comprising a memory and at least one processor, the memory storing a computer program and the processor executing the computer program to enable the electronic device to perform the method as described in any of the first aspects.

[0037] This application provides a method and system for evaluating the rock-breaking effect of roller cutters, which uses a closed-loop logic of multimodal data fusion and iterative optimization in the evaluation process of roller cutter rock-breaking effect.

[0038] The beneficial effects of the technical solution provided in this application are as follows: On the one hand, by using dual-modal cross-validation of point cloud information and image information, the geometric and mechanical characteristics of the rock mass can be accurately depicted, reducing the misjudgment rate caused by the heterogeneity of the rock mass, thereby improving the accuracy of the selection of the rock breaking area by the roller cutter; on the other hand, by using the overlap of the first and second recommended areas as a feedback indicator, and in conjunction with the online iterative update of the roller cutter parameters (e.g., selecting from a preset set of roller cutter parameters), parameter self-optimization can be achieved. Attached Figure Description

[0039] The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Throughout the drawings, the same reference numerals denote the same components. Obviously, the drawings described below are merely some embodiments described in this application, and those skilled in the art can obtain other drawings based on these drawings.

[0040] Figure 1 This is a flowchart illustrating a method for evaluating the rock-breaking effect of a roller cutter, as provided in an embodiment of this application.

[0041] Figure 2 This is a schematic diagram illustrating a process for acquiring point cloud information and image information, provided in an embodiment of this application.

[0042] Figure 3 This is a schematic diagram of a process for obtaining selected hobbing cutter parameters, provided as an embodiment of this application.

[0043] Figure 4 This is a schematic diagram of a process for selecting a first recommended region, provided as an embodiment of this application.

[0044] Figure 5 This is a schematic diagram of a process for obtaining job evaluation prediction values, provided as an embodiment of this application. Detailed Implementation

[0045] To enable those skilled in the art to better understand the technical solutions in the embodiments of this application, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. It should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application. Furthermore, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily obscuring the concepts disclosed in this application.

[0046] The technical solutions of the embodiments of this application and how the technical solutions of the embodiments of this application solve the above-mentioned technical problems will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that the various embodiments or technical features described below can be arbitrarily combined to form new embodiments. The order of description of the embodiments below is not intended to limit the preferred order of embodiments. The same or similar concepts or processes may not be described again in some embodiments. Obviously, the described embodiments are some embodiments of the embodiments of this application, but not all embodiments.

[0047] Method implementation examples.

[0048] See Figure 1This embodiment provides a method for evaluating the rock-breaking effect of a roller cutter. The method can run on cloud servers and edge servers for evaluating the effectiveness of rock-breaking tests. The method includes:

[0049] S101, acquire point cloud information and image information of multiple candidate working areas including the target rock working surface, wherein the size of the cutter placement area of ​​each candidate working area meets the spatial constraints of the roller cutter rock breaking operation; select a first recommended area from the multiple candidate working areas based on the point cloud information; and acquire a second recommended area from the multiple candidate working areas based on the image information and the selected roller cutter parameters.

[0050] S102, obtain the job evaluation prediction value based on the overlap between the first recommended region and the second recommended region. When the job evaluation prediction value is lower than the preset value, re-acquire the image information and select the cutting tool parameters, and update the second recommended region based on the new image information and the new selected cutting tool parameters. Use the first recommended region and the updated second recommended region to re-judge the job evaluation prediction value until the job evaluation prediction value is not lower than the preset value.

[0051] S103, take the predicted value of the operation evaluation when it is not lower than the preset value and the corresponding selected cutter parameters as rock breaking decision information; generate an operation strategy according to the selected cutter parameters in the rock breaking decision information and conduct a rock breaking test on the target rock; obtain inspection information including image data used to indicate the geometric characteristics of the rock debris after rock breaking, and obtain the evaluation result of cutter rock breaking according to the inspection information and the predicted value of the operation evaluation in the rock breaking decision information.

[0052] Steps S101 and S102 can be run on an edge server. In step S103, the edge server sends the inspection information to the cloud server and obtains the evaluation results fed back by the cloud server, forming a two-level edge-cloud collaborative architecture. The point cloud-image fusion and region / parameter iteration processes are completed at the edge, resulting in low network latency and meeting the needs of continuous testing. The edge side only uploads the final inspection information, while the original high-density point cloud and multi-view images remain in the edge cache or local disk, reducing uplink bandwidth and cloud storage pressure. In specific applications, the cloud is only responsible for the evaluation and archiving after the rock-breaking test, as well as the training and updating of the model involved in this application. It can be considered that the technical solution of this application embodiment realizes the roller rock-breaking test through the image synthesis calculation of the first recommended region and the second recommended region, and the closed-loop logic of multimodal data fusion-iterative optimization, for the evaluation process of the roller rock-breaking effect, and finally obtains the corresponding evaluation results.

[0053] As an example, three-dimensional geometric features of the working face (such as joint and fracture orientation, concavity and convexity, and roughness of exposed rock surfaces) are acquired using lidar or structured light scanning to quantify rock mass excavability. Two-dimensional texture / thermal radiation features (such as lithological color, microcrack distribution, and cutting temperature field) are acquired using industrial cameras or infrared thermal imagers to identify local weak interlayers or highly abrasive areas. Regions with high rock mass integrity and good geometric flatness are extracted using point cloud normal vector clustering algorithms (such as region growing) as priority areas for mechanical excavability. The images are input into a convolutional neural network (CNN) to predict the rock mass compressive strength, and combined with cutter parameters (such as cutter spacing S and penetration P), the model calculates and selects the regions with the highest matching degree with the selected cutter parameters. The Jaccard index is used to quantify the spatial overlap ratio of the first / second recommended regions (e.g., overlap ≥80% is the threshold). If it is below the threshold, it indicates that the geometric features and mechanical features do not match (e.g., the point cloud shows a complete rock mass, but the image identifies it as a fractured zone).

[0054] Simultaneously, a parameter self-correction mechanism can be triggered to obtain and adjust the hobbing parameters for the cutter spacing and / or penetration, recalculate the second recommended area, and continue until the overlap meets the standard. The final job evaluation prediction value and the corresponding hobbing parameters (such as the optimized cutter spacing S=85mm, penetration P=8mm / rev) are packaged and sent to the control terminal of the TBM testing equipment.

[0055] The above technical solution has the following advantages: On the one hand, by using dual-modal cross-validation of point cloud information and image information, the geometric and mechanical characteristics of the rock mass can be accurately depicted, reducing the misjudgment rate caused by the heterogeneity of the rock mass, thereby improving the accuracy of the selection of the rock breaking area by the cutter; on the other hand, by using the overlap of the first and second recommended areas as a feedback indicator, and in conjunction with the online iterative update of the cutter parameters (e.g., selecting from the preset cutter parameter set), parameter self-optimization can be completed.

[0056] See Figure 2 In one exemplary embodiment, acquiring point cloud information and image information of multiple candidate working areas including the target rock working face includes:

[0057] S201: Multiple line laser scanners are rotated and arranged in front of the target rock working area to acquire point cloud data covering each candidate working area;

[0058] S202 simultaneously uses multiple industrial cameras to capture images of various candidate work areas from different perspectives, performs registration and distortion correction on the obtained multi-view images, and generates image information.

[0059] S203, the point cloud data and the image information are spatiotemporally synchronized with the spatial calibration plate using timestamps to form a one-to-one point cloud-image data pair and stored in the database.

[0060] As an example, multiple line laser scanners are arranged on the same rotating support in front of the target rock working surface. The scanners are arranged in a circumferential array. During the rotation, the laser lines sweep across the entire working surface to be selected, generating a high-density three-dimensional point cloud. At the same time, multiple industrial cameras rigidly fixed to the scanners capture images synchronously from different perspectives, covering all texture, crack and temperature details.

[0061] Meanwhile, the multi-view images acquired by the industrial camera are first calibrated using a checkerboard or circular calibration plate to determine the intrinsic parameters (distortion coefficients) and extrinsic parameters (camera pose). Then, geometric registration is performed using SIFT / SURF feature point matching and RANSAC, and distortion correction is completed using the Brown-Conrady model. Finally, two-dimensional image information is output.

[0062] The scanner and camera can share the same trigger pulse, and the data is timestamped. A mapping relationship between the point cloud coordinate system and the image pixel coordinate system is established through a pre-deployed spatial calibration board, achieving spatiotemporal alignment frame by frame and point by point, generating one-to-one corresponding point cloud-image data pairs. The aligned data pairs can be compressed and stored in a database according to the acquisition batch for subsequent real-time retrieval.

[0063] The above technical solution offers the following advantages: Firstly, a single scan using a rotating linear laser array can obtain a complete 3D topography, avoiding multi-station stitching errors and improving point cloud density. Secondly, simultaneous acquisition by multi-view industrial cameras, followed by registration and distortion correction, ensures both image resolution and geometric accuracy, reducing false crack identification and texture distortion. Thirdly, the synchronization mechanism based on a unified spatiotemporal calibration plate with timestamps and spatial calibration ensures strict alignment between the point cloud and the image, allowing subsequent region recommendation models to directly call the corresponding data pairs without secondary registration, thus shortening computational latency. In summary, this simultaneous acquisition scheme, with high-density point clouds, high-precision multi-view images, and spatiotemporal alignment as its core, provides front-end data support for the evaluation of roller cutter rock breaking.

[0064] In an exemplary embodiment, obtaining a second recommended region from multiple candidate working regions based on the image information and selected hobbing parameters includes:

[0065] The image information and the selected hobbing cutter parameters are input into the target area recommendation model, and the candidate operation area corresponding to the output area range is used as the second recommended area.

[0066] The target region recommendation model is trained from a deep learning model, and the training process of the target region recommendation model includes the following steps:

[0067] Obtain a training set, which includes multiple sets of training data. Each set of training data includes sample image data, sample roller cutter parameters, and annotation data of the target region corresponding to the sample image data and the sample roller cutter parameters. For each set of training data in the training set, perform the following processing:

[0068] The sample image data and sample roller parameters in the training data are input into a preset deep learning model to obtain the prediction data of the corresponding target area.

[0069] Based on the obtained prediction data and labeled data, the model parameters of the deep learning model are updated; it is checked whether the preset training termination condition is met; if yes, the trained deep learning model is used as the target region recommendation model; if no, the deep learning model is trained again using the next training data.

[0070] A high-resolution image of the working surface to be tested, along with the currently selected hob's geometric / mechanical parameters (pitch, penetration, cutter head speed, etc.), are simultaneously input into the target region recommendation model. The target region recommendation model is a fully trained deep learning network (typically a CNN-Transformer hybrid network). It uses image tensors and hob parameter vectors as dual-channel inputs, and extracts the coupling features of rock surface texture, cracks, and hob cutting conditions through a convolution-attention mechanism. It outputs a recommended region mask, i.e., a high-efficiency region suitable for hob entry in a two-dimensional image coordinate system. The mask output by the model is then back-projected onto the actual candidate working area to obtain the second recommended region, completing the mapping from pixels to physical space.

[0071] The above technical solution offers the following advantages: Firstly, by simultaneously inputting the image and hob parameters into a deep learning network, the model can learn the nonlinear mapping relationship between subtle features of the rock surface and the cutting mechanics of the hob, thereby improving the accuracy of region selection from empirical levels to pixel levels, reducing the misjudgment rate caused by the heterogeneity of the rock mass. Secondly, the model only needs forward propagation to output recommended regions during the inference phase, resulting in a fast response time that meets the real-time control requirements of TBMs. Thirdly, the introduction of hob parameters as a conditional vector during the training phase allows for differentiated recommended regions to be generated for the same rock mass under different cutter spacings and penetration depths, enabling model reuse for the same rock mass under multiple cutter conditions and reducing data acquisition costs. In summary, this deep learning-driven second recommended region generation mechanism deeply integrates image perception, cutter conditions, and rock-breaking efficiency, achieving high intelligence while improving the accuracy of region selection.

[0072] See Figure 3 In one exemplary embodiment, the method for obtaining the selected hob parameters includes:

[0073] S301, Obtain the mapping relationship between preset image information and preset hob parameters; Based on the mapping relationship, obtain the preset hob parameters corresponding to the preset image information with the highest correlation to the image information, and use them as the selected hob parameters;

[0074] S302, when the predicted value of the job evaluation is lower than the preset value, the preset cutter parameter corresponding to the currently selected cutter parameter is masked from the mapping relationship, and the preset cutter parameter corresponding to the preset image information with the highest correlation with the image information is obtained according to the mapping relationship and used as the new selected cutter parameter.

[0075] In practical applications, a database of pre-set mapping relationships between preset image information and preset cutter parameters is pre-configured in the edge server. Each mapping relationship can come from previous experiments or historical data, representing the statistical correspondence between "rock surface texture / temperature / color characteristics → optimal cutter spacing, penetration depth, and cutterhead rotation speed". Real-time image information of the current working face is collected, and image hashing or deep feature extraction algorithms (such as the 2048-dimensional feature vector of ResNet-50) are used to calculate the similarity with the preset image information in the mapping database. The mapping record with the highest similarity is selected, and its corresponding preset cutter parameters are used as the selected cutter parameters.

[0076] When the predicted value of the job evaluation is lower than the preset threshold, it is automatically determined that "the currently selected cutter parameters are not suitable for the rock mass of the current area"; the mapping record is immediately blocked in the mapping library (marked as unavailable) to prevent subsequent loop calls; similarity matching is re-executed to select the suboptimal mapping record, and the new selected cutter parameters are output. The above blocking-reselection process is repeated until the predicted value meets the requirements or the mapping records in the library are exhausted.

[0077] The above technical solution has the following advantages: Firstly, it obtains the initial values ​​of the roller cutter parameters through the image-roller parameter mapping relationship, avoiding the time waste caused by traditional manual trial cutting; secondly, it introduces a masking mechanism to immediately remove unreasonable mappings when parameter mismatch is detected, preventing negative samples from being reused in the current closed-loop control and improving the overall recommendation accuracy; thirdly, the mapping library can be continuously supplemented with new samples as the project progresses, and masked records can be restored after manual review. In summary, this roller cutter parameter acquisition method, with its three-step closed-loop approach of pre-stored mapping + similarity retrieval + dynamic masking, balances real-time performance, accuracy, and scalability.

[0078] See Figure 4 In one exemplary embodiment, selecting a first recommended region from a plurality of candidate job regions based on the point cloud information includes:

[0079] S401, calculate the normal vector of the point cloud of each candidate work area, and extract the area with a local curvature change rate greater than a preset threshold as a candidate area of ​​high stress concentration.

[0080] S402, based on curvature similarity and spatial continuity, aggregates high stress concentration candidate areas into several continuous regions through a region growing algorithm, and outputs their boundary coordinates as the first recommended region.

[0081] As an example, the normal vector n is calculated point by point in the point cloud of each candidate work area, and the local curvature change rate κ = |∂n / ∂s| (where s is the arc length) is further calculated. When κ exceeds a preset threshold, the point is marked as a candidate point for high stress concentration, transforming the microscopic undulations of the rock mass surface into an indirect indicator of internal stress concentration. Considering that local micro-undulations (crests-troughs) on the rock mass surface generate stress concentration under the action of the cutter normal load, in continuum mechanics, curvature κ is proportional to the spatial change rate of the surface normal vector. Based on the above formula ∂, which represents the derivative along the minimum change, |∂n / ∂s| on the discrete point cloud is the magnitude of the derivative of the normal vector along the tangential direction, which has first-order equivalence with the principal curvature.

[0082] Using all candidate points as seeds, a region growing algorithm is executed based on the criteria that the curvature difference is no greater than a preset condition (e.g., 0.05m⁻¹) and that the points are spatially adjacent. The algorithm continuously merges adjacent points in space to form several spatially continuous high-stress concentration areas with consistent curvature characteristics, thereby eliminating isolated noise points and ensuring that the subsequent hobbing cutter can operate stably in a continuous region.

[0083] For each contiguous region after aggregation, the boundary is extracted, and the three-dimensional boundary coordinates (x, y, z point series) are output. This boundary can be directly imported into the TBM control system as the first recommended area for the next actual cutting by the cutter head.

[0084] The algorithm can be considered to expand layer by layer according to adjacency rules (such as eight-neighborhood / twenty-six-neighborhood). As long as a neighboring point meets the curvature threshold, it can be incorporated into the current region; otherwise, growth stops. Mathematically, this mechanism is equivalent to performing a morphological closing operation on a set of high-stress-concentration points in a three-dimensional discrete space, making the final region a simply connected continuous region. The continuous region obtained from the working face ensures that the cutting edge and the rock surface can maintain linear or surface contact during the cutting process, reducing impact loads caused by intermittent contact; the uniform curvature ensures that the rock mass strength is quasi-homogeneously distributed within the region, preventing instantaneous overload caused by local hard points.

[0085] The above technical solution has the following advantages: by utilizing the dual features of normal vector and curvature, macroscopic judgment is transformed into a geometric-mechanical mapping, improving the accuracy of high-stress area identification. The algorithm only involves normal vector, curvature, and adjacency judgment, does not require large computing power, and can be completed in edge computing units.

[0086] See Figure 5 In an exemplary embodiment, obtaining the job evaluation prediction value based on the overlap between the first recommended region and the second recommended region includes:

[0087] S501, Project the first recommended region and the second recommended region onto the same reference plane coordinate system respectively, and calculate the ratio of their intersection area to their union area as the degree of overlap;

[0088] S502, if the overlap is not less than its corresponding preset overlap, the operation evaluation prediction value is obtained based on the overlap and the roller rock breaking efficiency coefficient; if the overlap is less than its corresponding preset overlap, the operation evaluation prediction value is obtained based on the overlap, correction coefficient and roller rock breaking efficiency coefficient.

[0089] The efficiency coefficient of the cutter rock breaking is obtained by referring to a table from the cutter head rotation speed, penetration depth and uniaxial compressive strength of the rock in the cutter parameters.

[0090] In practical applications, the candidate work area corresponding to the union of the two is used as the selection of the target work area. When generating the work strategy based on the selected cutter parameters in the rock breaking decision information, the rock breaking operation corresponding to the work strategy is applied to the target work area. In this case, through dual-modal cross-validation of point cloud information and image information, the geometric and mechanical characteristics of the rock mass can be accurately characterized, reducing the misjudgment rate caused by the heterogeneity of the rock mass, thereby improving the accuracy of the cutter rock breaking area selection.

[0091] As an example, if the overlap is ≥0.8, the predicted value of the operation assessment is the product of the overlap and the efficiency coefficient of the roller cutter rock breaking; if the overlap is <0.8, the predicted value of the operation assessment is the product of the overlap and the efficiency coefficient of the roller cutter rock breaking, and then multiplied by the correction factor of 0.7.

[0092] The first recommended region (the high-stress complete area of ​​the point cloud) and the second recommended region (the image patch output by the deep learning model) in 3D space are projected onto the same 2D reference plane coordinate system through calibration matrices to ensure a one-to-one spatial correspondence between the two. The intersection area A∩ and the union area A∪ of the two regions are obtained on the reference plane, and the overlap ratio IoU = A∩ / A∪ is calculated.

[0093] If IoU ≥ 0.8: Predicted value of job evaluation = IoU × Cutter rock breaking efficiency coefficient η;

[0094] If IoU < 0.8: Job evaluation prediction value = IoU × η × 0.7 (correction coefficient). Where η is obtained in real time from the current cutter parameters (cutter head speed, penetration) and the rock uniaxial compressive strength UCS through an offline calibration table, reflecting the cutter-rock matching efficiency.

[0095] The above technical solution has the following advantages: Firstly, by using a unified coordinate system projection and IoU quantization, it achieves an objective measurement of spatial consistency between the two regions, avoiding misjudgments caused by coordinate deviations. Secondly, the threshold-based hierarchical processing retains high confidence in high-overlap scenarios while introducing correction coefficients for low-overlap scenarios, ensuring that the evaluation prediction value is always within a reasonable range and preventing overly aggressive decisions. Thirdly, the efficiency coefficient of the roller cutter rock breaking is obtained instantaneously using an offline table lookup method, balancing accuracy and real-time performance. Fourthly, the entire formula requires only multiplication and one table lookup, resulting in extremely low computational load, making it easy to complete in the edge computing unit of the TBM. In summary, this solution significantly improves the reliability and real-time response capability of the job evaluation prediction value while maintaining the simplicity of the algorithm.

[0096] In an exemplary embodiment, obtaining the evaluation result of the roller cutter rock breaking based on the inspection information and the operation evaluation prediction value in the rock breaking decision information includes:

[0097] Based on the inspection information, evaluation data is obtained through the rock breaking evaluation model; based on the evaluation data and the operation evaluation prediction value, the evaluation result of the roller cutter rock breaking is obtained.

[0098] As an example, after the rock-breaking test is completed, two types of data are input into the rock-breaking evaluation model. Image data is used to indicate the geometric characteristics of the rock fragments, and combined with weighing data, it is used to indicate the weight distribution. The predicted job evaluation value P obtained from the aforementioned closed-loop optimization is also obtained.

[0099] The model is used to segment rock fragment images, extracting the equivalent diameter dᵢ, aspect ratio aᵢ, and angularity coefficient cᵢ for each particle; simultaneously, the mass fraction wᵢ of each particle size is calculated using weighing data; {dᵢ, aᵢ, cᵢ, wᵢ} are combined into a feature vector F, which is then input into a pre-trained rock breaking evaluation model (e.g., a CNN+MLP hybrid structure). The rock breaking evaluation model is a neural network model, and the network outputs two evaluation data points: the measured rock breaking efficiency index E. real (Unit energy consumption crushing volume, m³ / kWh) and rock slag gradation uniformity coefficient K (values ​​range from 0 to 1, the closer to 1, the more uniform the gradation).

[0100] E real The weighted fusion formula for K and the predicted value of job evaluation P is: S = α·P + β·Normalize(E) real )+γ·K.

[0101] Where α+β+γ=1, and the weights are determined by previous calibration tests. S is the final evaluation result of the roller cutter rock breaking mechanism, which can be directly mapped to the tunneling speed vadv and the average rock breaking amount per cutter V0. single The prediction interval. Normalize(E) realThis refers to mapping the measured rock breaking energy efficiency index to the [0,1] interval through linear normalization, so that it is on the same dimension as the operation evaluation prediction value P and the gradation uniformity K, so as to achieve weighted fusion.

[0102] In an exemplary embodiment, obtaining the evaluation result of the roller cutter rock breaking based on the inspection information and the operation evaluation prediction value in the rock breaking decision information includes:

[0103] The operation evaluation prediction values ​​in the inspection information and the rock breaking decision information are sent to the user equipment as test information; the evaluation data of the test information sent by the user through the user equipment is obtained as the evaluation result.

[0104] Test information is transmitted to user devices (PCs, tablets, or mobile apps) via wired / wireless networks. Upon receiving the test information, the user device simultaneously displays rock fragment photos / 3D reconstruction images, corresponding inspection information, and a predicted operational assessment value (P) on a visual interface. Users can score the rock breaking effect based on their field experience or select preset levels to generate evaluation data. This evaluation data is then transmitted back to the edge server via the same network link, becoming the evaluation result.

[0105] In one exemplary complete embodiment, a method for evaluating the rock-breaking effect of a rolling cutter includes:

[0106] The edge server is configured to acquire point cloud data covering each candidate work area, the point cloud data being acquired using multiple line laser scanners arranged in a rotating manner in front of the target rock work area;

[0107] Simultaneously, multiple industrial cameras are used to capture images of each candidate work area from different perspectives to obtain shooting information. The obtained multi-view images are then registered and distortion corrected to generate image information.

[0108] The point cloud data and the image information are synchronized in time and space using timestamps and a spatial calibration board to form a one-to-one point cloud-image data pair and store it.

[0109] Based on the point cloud information, the normal vector of the point cloud of each candidate operation area is calculated, and the area with a local curvature change rate greater than a preset threshold is extracted as a candidate area for high stress concentration. Based on curvature similarity and spatial continuity, the candidate areas for high stress concentration are aggregated into several continuous areas by a region growing algorithm, and these areas are used as the first recommended areas, and their boundary coordinates are output.

[0110] The image information and the selected hobbing cutter parameters are input into the target area recommendation model, and the candidate operation area corresponding to the output area range is used as the second recommended area.

[0111] The first recommended region and the second recommended region are projected onto the same reference plane coordinate system, and the ratio of their intersection area to their union area is calculated as the degree of overlap. If the degree of overlap is not less than its corresponding preset degree of overlap, the predicted value of the operation assessment is obtained based on the degree of overlap and the rock breaking efficiency coefficient of the roller cutter. If the degree of overlap is less than its corresponding preset degree of overlap, the predicted value of the operation assessment is obtained based on the degree of overlap, the correction coefficient, and the rock breaking efficiency coefficient of the roller cutter.

[0112] When the predicted value of the job evaluation is lower than the preset value, the image information and the selected hobbing parameters are reacquired, and the second recommended area is updated according to the new image information and the new selected hobbing parameters; the predicted value of the job evaluation is re-judged using the first recommended area and the updated second recommended area until the predicted value of the job evaluation is not lower than the preset value.

[0113] The predicted operation evaluation value at a value not lower than a preset value and its corresponding selected cutter parameters are used as rock breaking decision information; an operation strategy is generated based on the selected cutter parameters in the rock breaking decision information, and a rock breaking test is conducted on the target rock. During the test, the cutter rock breaking is carried out in the candidate operation area where the union region is located; inspection information including image data used to indicate the geometric characteristics of the rock debris after rock breaking is obtained; evaluation data is obtained through the rock breaking evaluation model based on the inspection information; the evaluation data and the predicted operation evaluation value are pushed to the cloud server.

[0114] The cloud server is configured to train a deep learning model to obtain a target region recommendation model, and the trained target region model is deployed on each edge server;

[0115] When any edge server obtains the second recommended area, according to the request instruction of the corresponding edge server, the mapping relationship between the preset image information and the preset rolling cutter parameters is pushed to the edge server corresponding to the request instruction.

[0116] Based on the evaluation data pushed by any edge server and the operation evaluation prediction value, the evaluation result of the rolling cutter rock breaking is obtained and fed back to the edge server, forming a two-level edge-cloud collaborative architecture.

[0117] When the evaluation result is lower than a preset threshold (e.g., S < 0.6, or the evaluation is deemed unsatisfactory by staff via user device), a count is incremented by one. When the count value is higher than a predetermined value (e.g., 10, 15), a prompt message is generated and sent to the user device, prompting the user to adjust and update the mapping relationship between the preset image information and the preset roller parameters. Adjustment, for example, involves re-labeling the preset roller parameters for the preset image information. Update, for example, involves obtaining the selected roller parameters corresponding to the count value from each edge server, and then assigning a penalty factor between (0,1) to each parameter, reducing their ranking priority in subsequent similarity retrieval.

[0118] The target region recommendation model is trained from a deep learning model, and the training process of the target region recommendation model includes the following steps:

[0119] Obtain a training set, which includes multiple sets of training data. Each set of training data includes sample image data, sample roller cutter parameters, and annotation data of the target region corresponding to the sample image data and the sample roller cutter parameters. For each set of training data in the training set, perform the following processing:

[0120] The sample image data and sample roller parameters in the training data are input into a preset deep learning model to obtain the prediction data of the corresponding target area.

[0121] Based on the obtained prediction data and labeled data, the model parameters of the deep learning model are updated; it is checked whether the preset training termination condition is met; if yes, the trained deep learning model is used as the target region recommendation model; if no, the deep learning model is trained again using the next training data.

[0122] The methods for obtaining hob parameters include:

[0123] Obtain the mapping relationship between preset image information and preset cutting tool parameters from the cloud server; obtain the preset cutting tool parameters corresponding to the preset image information with the highest correlation with the image information according to the mapping relationship, and use them as the selected cutting tool parameters;

[0124] When the predicted value of the job evaluation is lower than the preset value, the preset cutter parameter corresponding to the currently selected cutter parameter is masked from the mapping relationship. The preset cutter parameter corresponding to the preset image information with the highest correlation with the image information is obtained according to the mapping relationship and used as the new selected cutter parameter.

[0125] As an example, the experimental objective is to determine the optimal cutter entry position, cutter spacing S, and penetration depth P within multiple candidate areas at the tunnel face, and to provide a final rock-breaking performance score. Specifically, this includes:

[0126] P1, the edge server performs multimodal synchronization information collection.

[0127] A laser array rotates and scans to acquire a high-density point cloud over the entire area, while a camera array simultaneously captures high-resolution images from multiple perspectives. Then, spatiotemporal calibration and alignment are performed to generate and store point cloud-image data pairs that correspond one-to-one.

[0128] P2, the first recommended region R1 is obtained. Candidate points with high stress concentration are marked by point cloud normal vector and curvature calculation. After region growth and aggregation, the high integrity and low curvature continuous region R1 is output.

[0129] P3, the second recommended region R2 is obtained. The initial hobbing parameters (S0, P0) are selected by image feature retrieval. The CNN-Transformer is used to infer and output a mask that matches the tool condition, and finally R2 is generated.

[0130] P4, the overlap-degree closed-loop optimization step, calculates the IoU between R1 and R2, and determines whether IoU ≥ threshold. If so, locks the final (S, P) and the predicted job evaluation value P. eval Otherwise, mask the current parameters, update (S,P), regenerate R2 and recalculate the IoU between R1 and R2.

[0131] P5, at the rock breaking test site, the roller cutter rock breaking is carried out in the candidate operation area where the final (S,P) is located; rock debris images and weighing data are collected in real time to form inspection information.

[0132] P6, Evaluation process: Extract geometric feature vectors and output energy efficiency index E through rock breaking evaluation model. real And the uniformity of gradation K. Weighted fusion P eval E real The K and K are used to generate a comprehensive score S.

[0133] P7, visual push notifications, to display ratings on user devices.

[0134] Therefore, the technical solution provided in this application addresses the issue that existing rock-breaking assessment methods rely solely on post-processing energy calculations, with tool spacing and penetration depth locked empirically. This leaves the tool unable to correct for sudden changes in local rock mass strength during testing. The technical solution provided in this example completes region determination and cutter parameter closed-loop iteration in the initial stages of the rock-breaking test. Specifically, the rotating linear laser array and industrial camera can share trigger pulses and be calibrated using a spatial calibration plate. Curvature change rate is derived from point-by-point normal vectors, and then region growth is performed using curvature tolerance and adjacency criteria to eliminate isolated noise points, ensuring geometric continuity and consistent stress concentration characteristics in the cutter entry area. The CNN-Transformer dual-channel input method, with parallel input of image tensors and cutter parameter vectors, allows the attention mechanism to learn texture-tool coupling; a pixel-level recommended mask is output in a single forward pass during the inference stage, resulting in better region selection accuracy. The point cloud-image raw data remains locally, while the edge-cloud collaborative assessment method allows for region-(tool) parameter closed-loop completion without uploading all raw data; the cloud can focus on model retraining, further improving the applicability of this method. Simultaneously, through dual-modal cross-validation of point cloud and image information, the geometric and mechanical characteristics of the rock mass can be accurately characterized. Furthermore, during subsequent rock breaking, the roller cutter can be selected to break the rock in the candidate working area where the union region is located, reducing the misjudgment rate caused by the heterogeneity of the rock mass and thus improving the accuracy of the roller cutter breaking area selection. Additionally, region-parameter optimization is completed before the roller cutter enters the machine, avoiding the time and tool wear caused by traditional trial cutting-stopping-readjustment. Centralized cloud training and real-time edge deployment, combined with dynamic degradation failure mapping using penalty factors, ensure the long-term reliability of the strategy.

[0135] System Implementation Example.

[0136] This embodiment provides an evaluation system for the rock-breaking effect of a roller cutter. Its specific implementation and the achieved technical effects are consistent with the embodiments described in the above-mentioned method embodiments, and some details will not be repeated here. It includes:

[0137] The information acquisition module is used to acquire point cloud information and image information of multiple candidate working areas including the target rock working face, wherein the size of the cutter placement area of ​​each candidate working area meets the spatial constraints of the roller cutter rock breaking operation; based on the point cloud information, a first recommended area is selected from the multiple candidate working areas; based on the image information and the selected roller cutter parameters, a second recommended area is acquired from the multiple candidate working areas.

[0138] The region assessment module is used to obtain the job assessment prediction value based on the overlap between the first recommended region and the second recommended region. When the job assessment prediction value is lower than the preset value, the image information and the selected hobbing parameters are re-acquired, and the second recommended region is updated based on the new image information and the new selected hobbing parameters. The job assessment prediction value is re-judged using the first recommended region and the updated second recommended region until the job assessment prediction value is not lower than the preset value.

[0139] The result acquisition module is used to take the operation evaluation prediction value when it is not lower than a preset value and its corresponding selected cutter parameters as rock breaking decision information; generate an operation strategy based on the selected cutter parameters in the rock breaking decision information and conduct a rock breaking test on the target rock; acquire inspection information including image data used to indicate the geometric characteristics of rock debris after rock breaking; and obtain the evaluation result of cutter rock breaking based on the inspection information and the operation evaluation prediction value in the rock breaking decision information.

[0140] In one exemplary embodiment, the information acquisition module includes:

[0141] The point cloud acquisition unit is used to rotate and deploy multiple line laser scanners in front of the target rock working face to acquire point cloud data covering each candidate working area;

[0142] The image acquisition unit is used to simultaneously use multiple industrial cameras to capture images of each candidate work area from different perspectives, register and correct the distortion of the acquired multi-view images, and generate image information.

[0143] The data synchronization unit is used to synchronize the point cloud data and the image information in time and space with the spatial calibration plate through timestamps, forming a one-to-one point cloud-image data pair and storing it in the database.

[0144] In an exemplary embodiment, obtaining a second recommended region from multiple candidate working regions based on the image information and selected hobbing parameters includes:

[0145] The image information and the selected hobbing cutter parameters are input into the target area recommendation model, and the candidate operation area corresponding to the output area range is used as the second recommended area.

[0146] The target region recommendation model is trained from a deep learning model, and the training process of the target region recommendation model includes the following steps:

[0147] Obtain a training set, which includes multiple sets of training data. Each set of training data includes sample image data, sample roller cutter parameters, and annotation data of the target region corresponding to the sample image data and the sample roller cutter parameters. For each set of training data in the training set, perform the following processing:

[0148] The sample image data and sample roller parameters in the training data are input into a preset deep learning model to obtain the prediction data of the corresponding target area.

[0149] Based on the obtained prediction data and labeled data, the model parameters of the deep learning model are updated; it is checked whether the preset training termination condition is met; if yes, the trained deep learning model is used as the target region recommendation model; if no, the deep learning model is trained again using the next training data.

[0150] In one exemplary embodiment, the method for selecting the hobbing cutter parameters includes:

[0151] Obtain the mapping relationship between preset image information and preset cutter parameters; based on the mapping relationship, obtain the preset cutter parameters corresponding to the preset image information with the highest correlation to the image information, and use them as the selected cutter parameters;

[0152] When the predicted value of the job evaluation is lower than the preset value, the preset cutter parameter corresponding to the currently selected cutter parameter is masked from the mapping relationship. The preset cutter parameter corresponding to the preset image information with the highest correlation with the image information is obtained according to the mapping relationship and used as the new selected cutter parameter.

[0153] In an exemplary embodiment, selecting a first recommended region from a plurality of candidate job regions based on the point cloud information includes:

[0154] For each candidate work area, the normal vector is calculated from the point cloud, and the area with a local curvature change rate greater than a preset threshold is extracted as a candidate area for high stress concentration.

[0155] Based on curvature similarity and spatial continuity, a region growing algorithm is used to aggregate candidate regions of high stress concentration into several continuous regions, which are then used as the first recommended regions, and their boundary coordinates are output.

[0156] In an exemplary embodiment, obtaining the job evaluation prediction value based on the overlap between the first recommended region and the second recommended region includes:

[0157] Project the first recommended region and the second recommended region onto the same reference plane coordinate system, and calculate the ratio of their intersection area to their union area as the degree of overlap;

[0158] If the overlap is not less than its corresponding preset overlap, the predicted value of the operation assessment is obtained based on the overlap and the rock breaking efficiency coefficient of the cutter head; if the overlap is less than its corresponding preset overlap, the predicted value of the operation assessment is obtained based on the overlap, the correction coefficient and the rock breaking efficiency coefficient of the cutter head; wherein, the rock breaking efficiency coefficient of the cutter head is obtained by referring to a table from the cutter head parameters, such as the cutter head speed, the penetration depth and the uniaxial compressive strength of the rock.

[0159] In an exemplary embodiment, obtaining the evaluation result of the roller cutter rock breaking based on the inspection information and the operation evaluation prediction value in the rock breaking decision information includes:

[0160] Based on the inspection information, evaluation data is obtained through the rock breaking evaluation model; based on the evaluation data and the operation evaluation prediction value, the evaluation result of the roller cutter rock breaking is obtained.

[0161] Equipment implementation example.

[0162] This embodiment provides an electronic device, the specific embodiment of which is consistent with the embodiment described in the above method embodiment and the technical effects achieved, and some contents will not be repeated.

[0163] The electronic device includes a memory and at least one processor, the memory storing a computer program, and the at least one processor being configured to execute the computer program to implement the method as described in any of the method embodiments.

[0164] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application.

Claims

1. A method of evaluating the rock breaking performance of a rolling cutter, characterized in that, The method comprises the following steps: acquiring point cloud information and image information of a plurality of candidate operation regions of a target rock operation surface; selecting a first recommended region from the plurality of candidate operation regions according to the point cloud information; acquiring a second recommended region from the plurality of candidate operation regions according to the image information and selected cutter parameters; acquiring an operation evaluation prediction value according to the coincidence degree of the first recommended region and the second recommended region, reacquiring image information and selected cutter parameters when the operation evaluation prediction value is lower than a preset value, and updating the second recommended region according to the new image information and the new selected cutter parameters; rejudging the operation evaluation prediction value by using the first recommended region and the updated second recommended region until the operation evaluation prediction value is not lower than the preset value; taking the operation evaluation prediction value when the operation evaluation prediction value is not lower than the preset value and the corresponding selected cutter parameters as rock breaking decision information, generating an operation strategy according to the selected cutter parameters in the rock breaking decision information, and performing a rock breaking test on the target rock; acquiring inspection information including image data for indicating the geometric characteristics of rock debris after rock breaking, and acquiring an evaluation result of cutter rock breaking according to the inspection information and the operation evaluation prediction value in the rock breaking decision information.

2. The evaluation method according to claim 1, characterized in that The method for acquiring point cloud information and image information of a plurality of candidate operation regions of a target rock operation surface comprises the following steps: rotating a plurality of line laser scanners in front of the target rock operation surface to acquire point cloud data covering each candidate operation region; synchronously using a plurality of industrial cameras to shoot each candidate operation region from different angles, registering and correcting the obtained multi-angle images, and generating image information; synchronizing the point cloud data and the image information in time and space through a time stamp and a space calibration plate, forming a one-to-one point cloud-image data pair and storing it in a database.

3. The evaluation method according to claim 1, characterized in that The method for acquiring a second recommended region from a plurality of candidate operation regions according to image information and selected cutter parameters comprises the following steps: inputting the image information and the selected cutter parameters into a target region recommendation model, and taking the output region range corresponding to the candidate operation region as the second recommended region; wherein the target region recommendation model is obtained by training a deep learning model, and the training process of the target region recommendation model comprises the following steps: acquiring a training set, the training set comprising a plurality of training data, each set of training data comprising sample image data, sample cutter parameters, and annotation data of a target region corresponding to the sample image data and the sample cutter parameters; for each set of training data in the training set, the following processing is performed: inputting the sample image data and the sample cutter parameters in the training data into a preset deep learning model to obtain prediction data of the corresponding target region; updating the model parameters of the deep learning model based on the obtained prediction data and the annotation data; detecting whether a preset training end condition is met; if yes, taking the trained deep learning model as the target region recommendation model; if no, continuing to train the deep learning model using the next training data.

4. The evaluation method according to claim 3, characterized in that The selected cutter parameter acquisition method comprises: obtain a mapping relationship between preset image information and preset hob parameters; obtain, according to the mapping relationship, a preset hob parameter corresponding to preset image information with the highest degree of association with the image information, and take the preset hob parameter as a selected hob parameter; when the work evaluation prediction value is lower than a preset value, shield the preset hob parameter corresponding to the current selected hob parameter from the mapping relationship, obtain, according to the mapping relationship, a preset hob parameter corresponding to preset image information with the highest degree of association with the image information, and take the preset hob parameter as a new selected hob parameter.

5. The evaluation method according to claim 1, characterized in that The selecting, according to the point cloud information, a first recommended region from the plurality of candidate work regions comprises: perform normal vector calculation on the point cloud of each candidate work region, and extract a region with a local curvature change rate greater than a preset threshold as a high stress concentration candidate region; based on curvature similarity and spatial continuity, aggregate the high stress concentration candidate region into a plurality of continuous regions through a region growing algorithm, and take the continuous regions as the first recommended region, and output boundary coordinates of the first recommended region.

6. The evaluation method according to claim 5, characterized in that The obtaining, according to the coincidence degree of the first recommended region and the second recommended region, a work evaluation prediction value comprises: project the first recommended region and the second recommended region onto the same reference plane coordinate system respectively, and calculate a ratio of an intersection area to a union area of the first recommended region and the second recommended region as the coincidence degree; if the coincidence degree is not less than a preset coincidence degree corresponding to the coincidence degree, obtain the work evaluation prediction value according to the coincidence degree and a hob rock breaking efficiency coefficient; if the coincidence degree is less than the preset coincidence degree corresponding to the coincidence degree, obtain the work evaluation prediction value according to the coincidence degree, a correction coefficient and the hob rock breaking efficiency coefficient.

7. The evaluation method according to claim 1, characterized in that The obtaining, according to the check information and the work evaluation prediction value in the rock breaking decision information, an evaluation result of hob rock breaking comprises: obtain evaluation data through a rock breaking evaluation model according to the check information; and obtain the evaluation result of hob rock breaking according to the evaluation data and the work evaluation prediction value.

8. A system for evaluating the rock breaking performance of a rolling cutter, characterized by comprise: an information acquisition module, configured to acquire point cloud information and image information of a plurality of candidate work regions of a target rock work surface, a size of a tool placement region of each candidate work region satisfying a spatial constraint of hob rock breaking work; and select a first recommended region from the plurality of candidate work regions according to the point cloud information; obtain a second recommended region from the plurality of candidate work regions according to the image information and a selected hob parameter; a region evaluation module, configured to obtain a work evaluation prediction value according to a coincidence degree of the first recommended region and the second recommended region; when the work evaluation prediction value is lower than a preset value, reacquire image information and a selected hob parameter, and update the second recommended region according to new image information and a new selected hob parameter; and rejudge the work evaluation prediction value by using the first recommended region and the updated second recommended region until the work evaluation prediction value is not lower than the preset value. The result obtaining module is configured to: take the operation evaluation prediction value and the corresponding selected hob parameter when the operation evaluation prediction value is not lower than the preset value as rock breaking decision information; generate an operation strategy according to the selected hob parameter in the rock breaking decision information and perform a rock breaking test on the target rock; obtain inspection information including image data used for indicating the geometric characteristics of rock debris after rock breaking; and obtain an evaluation result of hob rock breaking according to the inspection information and the operation evaluation prediction value in the rock breaking decision information.

9. The evaluation system of claim 8, wherein, The information obtaining module comprises: A point cloud obtaining unit is configured to: arrange multiple line laser scanners in a rotating manner in front of the target rock operation surface, and obtain point cloud data covering each selected operation area; An image obtaining unit is configured to: synchronously use multiple industrial cameras to take pictures of each selected operation area from different perspectives, register and correct the obtained multi-perspective images, and generate image information; A data synchronizing unit is configured to: synchronize the point cloud data and the image information in time and space through time stamps and a space calibration plate, form a one-to-one point cloud-image data pair, and store the pair into a database.

10. An electronic device, comprising: The electronic device comprises a memory and at least one processor, the memory stores a computer program, and the processor is configured to execute the computer program, so that the electronic device can implement the method in any one of claims 1-7.

Images